Michael Hayes

I’ve spoken to over 2 dozen MDs at PE firms I can confidently say that the arb of figuring out how to implement Vertical AI at portfolio companies is very real right now It will fundamentally change underwriting for those who can do it predictably and unlock generational returns. Most are aware they need to act. Very few have.

454

64 Comments

EXCLUSIVE: Under the covers of the $18m Thread investment round Harry Brelsford interviews Matt Linn, co-founder and COO of Thread, an AI service desk that automates non-billable resources on the service desk, making them billable and reducing service costs through agentic workflows. Linn shares his background from a 10-year career with an MSP in New York City. Thread is announcing a new round of fundraising at the Series A stage to accelerate R&D and build faster. The company emphasizes its unique brand and customer-centric approach, showcasing this through creative booths at events. For 2026, Thread plans to enhance product offerings with better context for agents, more auto resolutions, and zero-touch tickets.   WATCH THE PODCAST HERE   Matt Linn's Background and Introduction to Thread - Harry Brelsford introduces Matt Linn, co-founder and COO of Thread, a company described as a "Wall Street darling" in the MSP industry. - Matt Linn shares his background, mentioning his 10-year career with an MSP in New York City called RFA, which focused on hedge funds and private equity funds. - Matt explains his passion for solving industry problems and his decision to create Thread to enhance the MSP business model. - Harry Brelsford asks Matt to provide a concise elevator pitch for Thread. Thread's Elevator Pitch and AI Service Desk - Matt Linn describes Thread as an AI service desk that automates non-billable resources on the service desk, making them billable and reducing service costs through agentic workflows. - Thread aims to create a new business model for MSPs with better unit economics and competitiveness at scale. - Harry Brelsford inquires about any breaking news from Thread. - Matt announces a new round of fundraising at the Series A stage, aiming to accelerate R&D to build bigger, better, and faster than ever. Brand Recognition and Innovative Booths - Harry Brelsford comments on Thread's strong brand recognition and the positive feedback they receive. - Matt emphasizes the importance of Thread's radical and customer-centric approach, highlighting their unique and intentional marketing strategies. - Matt shares examples of Thread's creative booths at events, such as a carnival nightmare theme and a Candy Cane Lane booth, to illustrate their commitment to innovation and customer experience. - Harry recalls past experiences with Thread's themed events, including a party bus and various costumes, showcasing their creative marketing efforts. Future Plans and Final Thoughts - Harry Brelsford asks Matt about any upcoming plans for 2026. - Matt reveals that Thread will be investing heavily in product development, focusing on providing better context for agents, more auto resolutions, and zero-touch tickets. - Harry thanks Matt for his participation and mentions the positive feedback on the MSP forecast. - Matt concludes by reiterating Thread's commitment to innovation and improving service efficiency for MSPs. https://tel…

8

1 Comment

What are the best benchmarks for deep research? Particularly interested in long-form reports vs single questions and answers. As a holiday project, I've been expanding the scope of Grep beyond business deep due diligence and stumbling into interesting (encouraging) results. I've been using DeepResearch Bench. 100 PhD-level tasks across 22 fields, created by USTC and Metastone researchers. Tests long-form report quality with their RACE framework. Caveats: uses Gemini 2.5 for scoring (outdated), and some baselines are third-party evals by the researchers that may not reflect current performance. Humanity's Last Exam (HLE) - Created by Center for AI Safety + Scale AI. 2,500 expert-level questions designed to be the "last academic exam" for AI. Not really about deep research - it's single Q&A format testing knowledge/reasoning, not synthesis into reports. DeepSearchQA - Released by Google DeepMind (December 2025). 900 prompts across 17 fields testing multi-step information seeking. Good: Tests comprehensiveness and exhaustive answer generation. Bad: Still focused on answer lists vs analytical long-form reports - closer to retrieval than research synthesis. What am I missing? Looking for benchmarks that evaluate the full research workflow: planning, gathering, synthesizing, and producing analyst-grade reports. Not just "find the answer" but "write me something I'd pay a consultant for."

8

2 Comments

With agentic workflows and longer context windows, it's easy to throw everything at the model and hope that the model/agent is smart enough to 'figure it out'. But if your context is hallucinated by LLMs ("context poisoning"), overpowers the LLM's trained behaviour/knowledge ("context distraction"), confuses the LLM with too many options ("context confusion"), have parts of the context that contradict one another ("context clash"), it's going to lead to suboptimal results. The answer to fixing the failure modes is to work on cleaning up the context via RAG or some form of context pruning.

8

2 Comments

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

AI transparency is starting to shape system design. Recent developments in California’s AI transparency requirements are one more signal of where this is heading. Expectations around disclosure, consent, and traceability are increasingly translating into concrete system requirements rather than policy statements. The jurisdiction matters less than the pattern. As AI systems scale, regulators and stakeholders are focusing less on stated intent and more on whether transparency is embedded in system design, data flows, and operating controls, and whether evidence can be produced when it matters. This reflects what I have seen repeatedly in regulated deployments. Transparency cannot be layered on after the fact. Consent cannot sit only at the edge of the journey. It has to be operational. Organisations that treat this as an architectural consideration tend to scale with fewer surprises. Those that approach it primarily as a compliance exercise often discover the gap later. Similar signals are beginning to appear in other regulatory and operating contexts. #AI #FinancialServices #Governance

4